Fable/Mythos shutdown --ultimate in FAFO

Anthropic and OpenAI have spent years leaning into the "too dangerous to release" mythos (lol) of their models. The latest one got shutdown after Amazon (and other) researchers claimed they were able to jailbreak the model into giving accurate instructions to create biological weapons and cybersecurity attacks.

--
Freeman's got so much riz they need to call him Aura Parseghian

Post reply

372 views

I have a Max account and use Claude Code extensively, both at work and personally. Most of the time, I do not notice a meaningful difference from one model version to the next, whether it is Opus, Sonnet, GPT, or something else.

Fable felt different almost immediately. It seemed to plan ahead, check its own work, and understand what I was really trying to get at rather than just answering the literal prompt. The results were often 80% to 90% of what I was looking for, which is better than what I usually get from other models.

Some of what it did also caught me off guard. It appeared to work around safeguards that usually prevent bots or AI tools from scraping websites, and in one case seemed to access a Barchart.com API in a way that gave me free historical futures data I have not been able to get just by browsing the site.

I was honestly disappointed to see it go over the weekend. Even though it used more tokens per request, it seemed to get closer to the right answer faster, so it may have been more efficient overall. At the same time, I can now see that some of the value may have come from things happening behind the scenes that were questionable, including giving me access to data I probably should not have had.

Post reply

119 views

Given his history and the shenanigans he's already gotten up to with this administration, this all feels a lot like, well, shenanigans.

--
Sometimes I rhyme slow sometimes I rhyme quick.

Post reply

277 views

I have to grudgingly hand it to Musk that he seems to be making good AI/tech decisions.

--
Freeman's got so much riz they need to call him Aura Parseghian

Post reply

172 views

Its not a good tech/ AI decision or at least I dont think I would define it exactly like that ...

Its a good vulture capitalist decision. He's in a bubble with a company that is a mess that he merged with a couple other messes, so he's using his new financial capital to buy a market winner.

That's not good tech from a Space X or AI point of view imo, that's just good profiteering from his financial bubble position.

I dont necessary see any value to Cursor for this deal. Do you really wanna use a tool that Musk could demand his new engineers corrupt?

The value in this deal looks to me like a winner getting gobbled up by a financial bubble.

What am I missing?

Post reply

172 views

But he wants to run an AI company. The problem is that X/Twitter data is junk for training.

He just bought probably the best training data. Cursor is sitting on a treasure trove of the highest quality developer data, using both Anthropic and OpenAI services.

I think this one is a very big deal.

--
Freeman's got so much riz they need to call him Aura Parseghian

Post reply

143 views

I'm increasingly convinced the pairing of the harness and the model is very important to quality of experience and output. Cursor reads to me like buying a horse farm at the damn of the automobile.

--
Sometimes I rhyme slow sometimes I rhyme quick.

Post reply

106 views

It certainly isnt a good deal for cursor customers.

Its just a financial play. Musk, once again, is going to leverage his newly gained assets to buy something that’s actually of great value, but has very little to do with his conglomeration of various Musk owned businesses.

And, if you look at the way Musk has destroyed/ manipulated Twitter … that’s going to be a disaster for cursor techies. How much has Musk’s terrible decisions driven customers/ users/ eyeballs away from twitter? He literally destroyed the twitter feed’s algorithm trying to promote himself and other wacko conspiracies. I expect the same disaster management to happen with Cursor.

Musk isnt bringing better value to the marketplace with this deal. He’s just using his financial leverage to buy something that’s actually good tech/ that he hopes will bring him more value, possibly just from a marketing perspective.

Post reply

134 views

Roughly 30% of the internet uses it monthly, and it's arguably the most politically influential 30%. Engagement is not at TikTok / YouTube levels but it's a more interactive medium for most users. MAU is generally up from when Musk bought it while bot count is reportedly down (not sure how that's going in the age of AI slop, though). Plus, it turns a reasonable profit.

Not sure it's delivered on the "open town square" that was promised, but it's hardly dying.

https://www.demandsage.com/twitter-statistics/

https://resourcera.com/data/social/x-users/

Post reply

134 views

Its clearly down significantly (25-50%) from its 2020-2022 numbers.

Half had ad revenue?

Its overall user numbers may have returned to some degree, but how much of that is RW bots and foreign fake accounts? It clearly hasnt recovered its US user numbers. (See “X users by region 2017 to 2024“ table in link below) … and that’s from company released information, so who knows how accurate it is … it could be much worse than published.

I dont think its anywhere near as productive of a place as it was pre-Musk and I have zero interest in wasting any time there nowadays. There are much better/ more accurate sources of near real time information on other platforms. I dont think its a well run business and I feel bad for any of my old colleagues that still work there (although I only worked there for a short project back a decade ago).

Besides why would you use a platform where you KNOW Elon Musk is openly corrupting the algorithm and spreading misinformation on purpose? Seriously? The guy in charge is blatantly lying to people and trying to manipulate public perception on that platform. I’m skeptical of anything from that platform going forward because their leadership is not even close to honest or honorable.

https://www.businessofapps.com/data/twitter-statistics/

Post reply

73 views

The real endgame is that everybody else wants OpenAI and Anthropic to lose.

SpaceX's reported Cursor acquisition made me wonder whether the most valuable asset in AI isn't the model at all. It's the data, workflows, feedback loops, and talent behind the model.

Imagine a world where:
• Frontier labs spend hundreds of billions racing for scale.
• Margins remain elusive because compute and infrastructure costs stay enormous.
• Public markets eventually demand profitability.
• Some labs stumble under the weight of their own capital requirements.

At that point, what are the truly valuable assets? Not necessarily the current model weights. It's

datasets
developer workflows
human feedback loops
enterprise relationships
engineering talent

In that scenario, the Mag 7 might not be trying to beat OpenAI and Anthropic. They might simply be waiting. Why spend years and billions building everything yourself if distressed assets eventually become available at a fraction of their peak valuations? Viewed through that lens, acquisitions like Cursor start looking less like software acquisitions and more like data acquisitions.

The most conspiracy-theory version of this idea goes even further: every regulatory challenge, lawsuit, or government action that increases the burn rate of smaller AI players becomes another banana peel on the path toward consolidation.

I don't actually believe there's a coordinated plan behind the scenes. But I do think it's possible that the eventual winners in AI are not the companies training today's most advanced models—they're the companies with the balance sheets to buy the best data, talent, and workflows if the current leaders ever stumble.

--
Freeman's got so much riz they need to call him Aura Parseghian

Post reply

121 views

And I'd say they've earned the benefit of the doubt on their strategic decisions.

Agreed the value long-term is the data, not the models. There are a limited number of customers who will pay Altman $200k/yr for frontier models. And everything else on the model end will race to the bottom, where open source is waiting patiently.

But data is defensible.

Which is where X/Twitter has one advantage... the best near-real-time textual data signals available, other than maybe WhatsApp or WeChat. I agree with Mark that it's high noise-to-signal and probably useless for things like coding development. But it is certainly useful in retail customer-facing situations. Which is why X is moving into banking and advertising, and Google is investing heavily in the same space and has YouTube data to work with, which is somewhat slower but maybe less noisy.

Meta has WhatsApp but I don't think has the financial horsepower to keep up on the compute, but we'll see. X may just rent compute power to everyone else on the side; it's how they handle SpaceX.

China and WeChat et.al. won't have access to the compute horsepower for some time, even with WeChat and TikTok providing good real-time data.

Post reply

114 views

Is that the mag 7 see the writing on the wall -- Gen AI doesn't have a viable business model at the moment -- and are looking for stuff like this as an off-ramp.

--
Freeman's got so much riz they need to call him Aura Parseghian

Post reply

193 views

and they are trying to create legends around, "our models are so strong" ... blah blah blah.

Maybe its true, but there's still a ton of hype w these companies and I definitely dont believe all of it ... let alone much of it.

In other news I saw a video today with a retirement planner reviewing the plan put together with the latest model from ChatGPT. Lots of good banter, but the model was trash when it came to actually calculating correct distributions and planning. Do NOT blindly trust an AI for calculations/stuff like that ... you gotta do your own homework.

The video then showed how the model really struggled with certain "predictions" like "how many "E's" are in the word seventeen? After 6+ voice queries it never once got it right.

"Lets be careful out there."
[image]

Post reply

254 views

the Polo Grounds

Fable/Mythos shutdown --ultimate in FAFO

I used fable for a few days and it was noticeably 'smarter'

Smells strongly of Sam Altman

SpaceX is buying Cursor

Really?

I can't speak to the finances

Eh.

Its a big deal, but I dont see it as a brilliant tech play

Twitter's demise is overrated

Tuh mae toe … tuh mah toe

here's what I think

That's certainly Apple's position.

my read

I heard one theory that this is just pre-IPO banter